[libc-commits] [libc] [libc][math] Implement an integer-only version of double precision sin with 1 ULP errors. (PR #184752)

[via libc-commits]

================
@@ -7,10 +7,21 @@
 //===----------------------------------------------------------------------===//
 
 #include "src/math/sin.h"
+#include "src/__support/macros/optimization.h"
+#include "src/__support/macros/properties/cpu_features.h"
 #include "src/__support/math/sin.h"
+#include "src/__support/math/sin_integer_eval.h"
 
 namespace LIBC_NAMESPACE_DECL {
 
-LLVM_LIBC_FUNCTION(double, sin, (double x)) { return math::sin(x); }
+LLVM_LIBC_FUNCTION(double, sin, (double x)) {
+#if defined(LIBC_MATH_HAS_SKIP_ACCURATE_PASS) &&                               \
+    defined(LIBC_MATH_SMALL_TABLES) &&                                         \
+    !defined(LIBC_TARGET_CPU_HAS_FPU_DOUBLE)
+  return math::integer_only::sin(x);
+#else
+  return math::sin(x);
+#endif
----------------
lntue wrote:

It will be, when we make the integer_only version correctly rounded.  At the moment, it still has 1 ULP errors.

https://github.com/llvm/llvm-project/pull/184752